An index of AI health just fell into a bear market

AN INDEX OF AI HEALTH JUST FELL INTO A BEAR MARKET

While some - starting with SoftBank boss Masayoshi Son - may consider it "blasphemy" to call the AI trade a bubble, markets, more often than not, will offer a reality check for that kind of fervour.

Semiconductor stocks around the world are still up massively so far this year, but they have taken a beating recently. The Philadelphia Semiconductor index NASDAQ:SOX has lost 12% in two weeks, while South Korea's Kospi index KRX:KOSPI is 14% below last month's record highs - although it is still worth nearly double what it was at the start of the year.

A near-$700 billion splurge by the hyperscalers on building the data centres to power the AI revolution is seen as a boon to the underlying economy and to markets, but the concern about when and how both the companies doing the build-out and those adopting the technology will start to see some returns.

One way of tracking usage and revenue is the price of tokens. Tokens are basically units of data that measure the amount of work an AI model does when it processes a request. AI usage is exploding and companies everywhere have generally encouraged heavy use of these tools, known as "tokenmaxxing".

AI companies are moving from flat subscriptions to usage-based pricing, which makes it even harder for companies with more unpredictable bills that can end up higher than expected. It isn't just proactive use of AI tools either. Models and apps that might run in the background can use huge amounts of tokens.

Silicon Data's LLM token Expenditure Index is a daily reference for token prices. The company describes it as "an expenditure or usage-weighted average token price index. It tells you how much currently the entire market AI is paying for a million LLM tokens irrespective of models."

The index is around $1.63 per million tokens, down nearly 20% from record highs above $2 in May - technically the definition of entry into a bear market and back roughly where it was two months ago, according to Citadel Securities, citing data from Silicon Data. Silicon Data itself says this decline in the index isn't necessarily evidence of a drop in AI demand, but rather, a sign that users may be getting pickier about what they'll pay for.